Home | About Us | Contacts | Help | Feedback | SEARCH Global | Australia | NSW | Vic. | Qld | WA | SA | Tas. | ACT | NT | Ant. | Weather & Warnings | Hydrology | Climate | Numerical Prediction | About Services | Learn About Meteorology | Registered User Services |

Global Guide
Main Menu

Global Guide to Tropical Cyclone Forecasting:
CHAPTER 3: TROPICAL CYCLONE MOTION

3.3 MOTION FORECASTING TECHNIQUES

The characteristics, underlying assumptions and methodology for currently used track forecasting techniques are described in this section. The techniques are loosely grouped into the following categories based on their major characteristics: Persistence, Climatology, Synoptic Techniques, Satellite Techniques, Expert Systems, Dynamical Methods, and Statistical-Dynamical Techniques.

3.3.1 Persistence

Persistence of the current track is easy to apply and has considerable skill for short-term forecasts. The simplest method is to use the current, or immediate past motion by maintaining either a constant direction and speed or constant meridional and zonal components. Because of the uncertainties inherent in analysing warning locations and analysis trends, a conservative approach of using the past motion over at least 6-12 hours instead of the current motion is recommended. The precise period to be used depends on the degree of smoothing that is required, or considered desirable.

The next level of complexity is to maintain the recent trend, or pattern in the analysis track. This can be achieved manually by maintaining the current curve in the track, but it is preferable to use some objective approach. Fitting a function to the track using least-squares is one such approach, but care needs to be taken with high-order polynomials, which can be quite unstable when used to extrapolate beyond the original data. An alternative method is to use a form of regressive filter, such as a Kalmen filter (Titus and Jarrell, 1985).

3.3.2 Climatology

Observations that tropical cyclones in certain regions and seasons move along similar tracks have given rise to a range of climatological techniques for forecasting motion. These techniques may be loosely grouped into three categories: climatological mean motion, analogue, and Markov chain.

The simplest approach is to calculate the mean motion of all tropical cyclones in a few degrees latitude/longitude square⁽¹⁾ for a specified time period. One method is to use means and standard deviations of meridional and zonal motion for 5^o Marsden squares by calendar month; the standard deviations give an indicator of the uncertainty in the climatological forecast. A variation on this method is to develop histograms for specified direction and speed ranges and to present these as cyclone roses on charts for display in the forecast office (Fig. 3.6, Lourensz, 1981).

Figure 3.6: Histograms of cyclone direction and speed, presented in rose format, for the Australian region in January (after Lourensz, 1981).

The climatological tracks can be used to calculate the probability of a tropical cyclone being in a specified area at each of the standard forecast time periods (Crutcher and Hoxit, 1974). A very effective variation on this method is to generate a scatter plot of the 12, 24, 48 and 72 hour positions of cyclones that moved through the current storm location (J. Martin, personal communication, 1989). This plot may contain all tropical cyclones or some subset, such as those within a certain time or motion window for the current storm. Such plots effectively indicate potential bifurcation regions where storms have a tendency to move in one of two (or more) preferred directions. This knowledge is useful for identifying potential for large forecast errors and for interpreting other techniques, such as CLIPER.

Analogue methods consist of searching for all tropical cyclones in the historical record that occurred within a preset spatial, seasonal and translational velocity range of the cyclone to be forecast (Hope and Neumann, 1971). The forecast track is derived as the mean of all these cyclones. Such analogue techniques are based on the assumption that tropical cyclones moving similarly to the current system will be in similar environments, so that their future movements will provide a good forecast. The scatter in analogue tracks provides an objective indicator of the uncertainty in the forecast. Operational analogue techniques are summarised in Table 3.2, the RECR technique is based only on a data base of recurving cyclones. Minor variations to these techniques used in Typhoon Committee countries are summarised in WMO (1987).

The Markov chain approach provides transition probabilities between speed and direction ranges, known as "bins", based on the climatological characteristics of previous cyclones in the region and time of year (Leslie et al., 1992). The selection of bins is constrained in that each must contain a similar proportion of cyclones with sufficient numbers to allow a stable prediction.

Of the above climatological techniques, we recommend the use of monthly charts listing roses of cyclone direction and movement (Fig. 3.6) to provide a very useful familiarisation with the monthly climatology for the region. The Markov chain technique provides useful indications of the likelihood of changes in speed and direction and thus the degree of consideration given to persistence. Analogue techniques are not recommended for primary track forecasting aids as they do not perform consistently better than other methods, and require substantial computing facilities to search the archive for track analogues. However, a variation on this approach, used by the JTWC (J. Martin, personal communication, 1990), is to keep all cyclone tracks in a data base on a workstation or personal computer, then to search for analogues by any criteria specified by the forecaster. This method provides a degree of flexibility in the analogue selection criteria and provides a ready, objective answer to the types of climatological questions that often arise in forecast discussions.

3.3.3 Combined Persistence and Climatology

A combination of persistence and climatology provides the best basic forecast technique and the basis for comparing forecasts from different regions. The first such technique was to combine persistence and climatology with equal weights, the ½(P+C) method developed at the Royal Observatory, Hong Kong (Bell, 1963; Chin, 1970). This is still used in a number of forecast offices (WMO, 1987) and is designated HPAC at JTWC. Operational analogue techniques also use persistence with decreasing weight over the first 36 hours.

In recent years the CLIPER technique (CLImatology and PERsistence, Neumann, 1972) has gained widespread acceptance (Leftwich and Neumann, 1977; Aoki, 1979; Xu and Neumann, 1985; WMO, 1987; Leslie et al. 1990). CLIPER consists of regression equations to predict the zonal and meridional displacement of a tropical cyclone for set time periods, usually 12, 24, 36, 48 and 72 hours. The equations are developed by some form of stepwise regression with the set of predictors including current and past position, intensity, motion, and Julian date. Up to three combinations of these predictors may be used. The selection criteria are based on the amount of variance explained from the climatological cyclone record and additional methods are used to further stabilise and enhance the final result (see Neumann, 1972). As a general outcome, persistence predictors are given the highest weighting for the first 24 hours or so, and climatology dominates at longer time periods.

An alternative approach has been developed within the Australian Bureau of Meteorology, in which forecasts are made in 12 hour time steps, with the previous forecast(s) being used as predictors (Morrison, Woodcock and Holland, personal communication, 1992). This provides improved forecasts of recurvature compared to the original CLIPER, but has not as yet been utilised widely.

We strongly recommend that all forecast offices use CLIPER as their basic forecast tool. Although a large computer is needed to develop the original equations, the forecasts can be made on any computing system, in essentially zero time. CLIPER also provides a convenient and consistent benchmark for indicating the skill of other forecast techniques, as is elaborated in Section 3.5.

3.3.4 Determining the Steering Current

Techniques to estimate the "steering current" in which a tropical cyclone is embedded have arisen from the notion that tropical cyclones are equivalent to corks in a stream, and that an accurate determination of the stream flow will provide excellent forecasts. Research over the past decade has shown that this simplistic picture is inaccurate and that interaction between the tropical cyclone and its environment has a marked impact on motion (e.g., Fiorino and Elsberry, 1989b; Shapiro and Ooyama, 1990; Smith et al., 1990; Wang and Holland, 1993, Holland and Wang, 1993). Nevertheless, up to 80% of the variance of tropical cyclone motion can be explained by the large scale environmental flow (Neumann, 1979; Keenan, 1982) and its estimation provides valuable support for track prediction. Operational techniques for determining an appropriate steering flow include space mean approaches, such as the MUSIC technique (MUltilevel Steering by Integrated Current) used in several Asian countries (WMO, 1987), and the Control Point method of Chin (1970). The major difficulties arise from removal of the cyclone circulation and in determining the appropriate level or layer mean.

George and Gray (1976), Chan and Gray (1982) and Holland (1984a) examined the relationship between tropical cyclone motion and the environmental flow using composite cyclones. They estimated the basic current by averaging over a 1-7^o lat radial band surrounding the cyclone and found significant and consistent track deviations, which agreed with those from composites of operational analyses by Brand et al. (1981). In the mean, cyclones tend to move polewards and westwards of this basic current. For example, low latitude, westward-tracking cyclones move faster and slightly poleward of the basic current, those moving to the northeast move slower and to the west. These findings are in qualitative agreement with the theoretical studies referenced in the previous paragraph (see also Carr and Elsberry, 1990). The theoretical studies also support forecaster experience that very large cyclones tend to move more independently of the environmental flow than do small systems.

Some studies have used geopotential heights to estimate the environmental flow (e.g., Chin, 1972) and George and Gray (1976) showed that these produced similar results to direct wind observations. However, height fields are noisy and unreliable in the deep tropics (Neumann, 1979) and modern wind analysis methods are becoming quite accurate. We recommend use of winds directly in determining the environmental flow surrounding tropical cyclones.

Removal of a symmetric cyclone is not recommended for operational applications, as this does not always provide a good indicator of the environmental flow. The cyclone may be asymmetric and non-linear asymmetries in the environment will be partially removed, as has been shown by the idealised studies of Evans et al. (1991) and Smith et al. (1990). A much better, but not perfect, approach is to use an appropriate filter to remove the cyclone scales, preferably only in the vicinity of the cyclone. A Fourier filter has been used with success by Marks (1989) and Ueno (1989), and Elsberry and Bohner (1993) have suggested a method of filtering consistent with known theoretical studies of tropical cyclone motion.

The appropriate level or layer mean to be used to indicate the steering current has been debated widely (Elsberry, 1988). The most consistent observational result is that the 700 or 500 hPa level provides the closest approximation (George and Gray, 1976; Neumann, 1979; Pike, 1985). This is supported by theoretical studies of Wang and Holland (1993) and Holland and Wang (1993). A more consistent and stable relationship is found for layer means, however. Holland (1984) argued against the inclusion of outflow and inflow layers and recommended the layer from 850-300 hPa. The layer from 900-500 hPa was found to be the best by Wang and Holland (1993)and Holland and Wang (1993). This agreed with the barotropic forecast studies of Velden and Leslie (1991) and Velden (1993), who also found a distinct relationship between intensity and depth of the optimum layer mean (Fig. 3.7). Many forecast offices consider that deeper layer-means, eg 850-200 hPa, are best for forecasting. This apparent conflict with observational and theoretical results probably arises from the lack of good mid-level observations in the vicinity of most tropical cyclones. In these cases, the inclusion of 200 hPa analyses, with their observations from satellite cloud-drift winds and commercial aircraft may provide a more stable analysis.

Figure 3.7: Optimal depth of the deep-layer mean flow (based at 850 hPa) used in barotropic model forecasts of tropical cyclone motion in the North Atlantic and Australian regions (after Velden, 1990).

Recommendation: Where possible, we recommend that charts of 850-500 or 850-300 hPa layer mean flow be utilised as a general forecast aid. Mass weighting must be used in calculating this layer mean. Where mid-level observations are poor, inclusion of the 200 hPa level in the layer mean may be beneficial. For strongly sheared systems, a lower mean, possibly just the 850 hPa level, should be considered. Fourier, or similar filtering of the flow to remove the cyclone scales is desirable.

3.3.5 Synoptic Methods

Many synoptic methods have been developed on the forecast bench to provide guidance in a number of difficult situations. Unfortunately, most of these are never verified and remain in the realm of forecast folklore. Although such methods can contain the combined experience of many years, they also can be demonstrably incorrect, hence only those methods that have been through some form of independent evaluation are discussed.

Recommendation: It is good forecast practice to keep a comprehensive log of the forecast decision process, and to use it in post-operational verifications. Such practice over a period of years can help to show inconsistencies in "favourite" but suspect forecast lore. It also forms the basis for development of check sheets and expert systems to enable a more coherent forecast process.

Figure 3.8: Synoptic situations associated with: a) rapidly accelerating cyclones, and b) those that move steadily down the Western Australian coast (Hanstrum and Foley, personal communication, 1988).

The most widespread synoptic pattern recognition involves the relative locations of the cyclone with the subtropical ridge and upper-level troughs in the mid-latitude westerlies. Generally, though not always, a strong subtropical ridge located poleward of the cyclone will result in continued movement westward. Similarly, presence of a westerly trough in the upper levels west of the cyclone is an excellent indicator of recurvature (Dunn and Miller, 1964; George and Gray, 1976). George and Gray particularly show that consistent increased troughing in the upper troposphere westerlies west and poleward of the cyclone can be observed 2-3 days prior to recurvature.

Hanstrum and Foley (personal communication, 1988) have developed a synoptic classification to discriminate cyclones that rapidly accelerate along and across the Western Australian coast and that have been responsible for major forecast failures. They stratified cyclones moving down the coast into two groups of rapidly and steadily moving systems, then composited the surface and upper-air fields to produce the synoptic typing shown in Fig. 3.8. This stratification has since successfully differentiated all rapidly accelerating cyclones, and provided improved forecast confidence and accuracy. Such synoptic classifications provide valuable guidance during periods of potentially major forecast failures and are strongly recommended.

Quasi-stationary seasonal features also can provide an indicator of the net movement of tropical cyclones, even though such features may be partially obscured on daily analyses. For example, the upper circulation over Asia is generally dominated by the Tibetan high during the summer monsoon and cyclones approaching China are prone to recurvature. When the Tibetan high is weakened, or unseasonably eastward, increased incidence of westward moving cyclones can be expected (Chen and Ding, 1979). Chen and Ding also describe how a quasi-stationary blocking high with attendant highly meridional flow can produce persistent track anomalies until it breaks down. Similarly, Holland and Pan (1979) showed quite strong relationships between cyclone tracks and the 30 day mean flow in the upper-troposphere over the Australian region.

At a smaller scale, interactions between adjacent weather systems can result in consistent track patterns that have a degree of forecastability, provided that the interactions are recognised. The most famous of these is the interaction of binary cyclones, known as the Fujiwhara effect (Fujiwhara, 1921, 1923, 1931). When two cyclones occur within 1500 km of each other they may interact in a manner that seems chaotic, but is made up of a combination of mutual rotation and advection by the larger scales (Brand, 1970). The basic analysis method (Fig. 3.9) is to plot the cyclones on a separate screen or map in a coordinate system that moves with the geometric centre of the line joining their centres. For cyclones which vary considerably in size, a weighted centre, which is closer to the larger cyclone, may be preferred, but generally the uncertainties are sufficient for us to recommend use of a simple geometric centre. Whilst the mutual interaction occurs, the forecast method is to separately forecast both the binary orbit (usually by persistence) and the movement of the system centre, then to add the two forecasts back together to obtain the actual forecasts.

Mutual cyclonic rotation of binary cyclones becomes apparent once they approach to within 1500 km (Brand, 1970) and its frequency of occurrence increases as the separation distance decreases (Fig. 3.10, Dong and Neumann, 1983). This graduated response of tropical cyclones to each other is related to the effects of the imposed environment. For example, an environment with a strong cyclonic (anticyclonic) horizontal shear will enhance (weaken) the binary cyclone rotation. Further, large tropical cyclones will exert an influence over a larger distance than will small systems.

Figure 3.9: Illustration of the technique used to separate the motion of binary tropical cyclones into mutual rotation about a common centroid (inset) and movement of the centroid. Dots indicate 6 h positions.

Figure 3.10: Proportion of binary tropical cyclones that undergo mutual cyclonic rotation as a function of separation distance (Dong and Neumann, 1983).

The global occurrence of binary tropical cyclone interaction is indicated by Fig. 3.11 (Neumann, personal communication, 1993). This figure provides a frequency graph of the number of simultaneous tropical cyclones within separation distances of up to 1850 km (1000 nm). Binary tropical cyclones are most common in the western North Pacific, where about 4 pairs occur each year within the 1500 km distance at which mutual interaction occurs. Binary interaction is also quite common in the eastern North Pacific basin and Australian/southwest Pacific regions. Indeed, if the number of interactions is normalised by tropical cyclone frequency⁽²⁾ (inset to Fig. 3.11), the rate of binary interaction is about the same for all three basins. However, as pointed out by Dong and Neumann (1983), binary tropical cyclones over the western North Pacific basin typically occur in the monsoon trough where large-scale steering currents are weak and binary interaction dominates. In the eastern North Pacific, and to a lesser extend the Australia/southwest Pacific basin, binary interaction often is masked by the stronger environmental steering.

Figure 3.11: Frequency of binary interaction over each of the ocean basins defined in Chapter 1 (Neumann, personal communication, 1993). The lower inset contains the average number of tropical cyclones per season for the specified time period.

Brand (1970), Bao et al. (1979), Dong and Neumann (1983) and Elsberry (1988) discuss operational aspects of binary tropical cyclones. A recent study by Lander and Holland (1993) has shown that binary cyclones do not follow the classical Fujiwhara model of mutual rotation and approach. Their revised model (Fig. 3.12) consists of a mutual approach, which tends to be more often anticyclonic than cyclonic. A period of rapid change during which the cyclones capture each other, followed by a long period of mutual rotation, during which the cyclones may or may not approach each other. The cessation of the interaction occurs either by destruction of one of the systems, which merges into the circulation of the dominant cyclone, or by rapid escape from each other's influence. In terms of chaos theory, the approach/escape, mutual orbit, and merged state are stable attractors which are separated by the transition regimes of capture, merger and escape.

Some of the details of what happens in these stages has been elaborated by Ritchie and Holland (1993) and by Holland and Dietachmayer (1993). Of importance to forecasting is their conclusion that the transitions are difficult, if not impossible to forecast, and that the dominant cyclone may continue to meander about a mean track, even after the secondary system has been sheared beyond the resolution of the observing system. Forecasters should realise, however, that capture and escape occur abruptly, so that actual forecast skill can be obtained during the orbit phase.

Figure 3.12: Model of binary tropical cyclone interaction developed by Lander and Holland (1993). See text for description of components. The tracks are plotted in a coordinate system relative to their mutual centre and each dot represents a 6 hour period.

Other mesoscale systems within the cyclone circulation can be expected to cause track meanders. Holland and Lander (1993) have shown that mesoscale convective systems (MCS) produce major track deviations, such as those for Typhoon Sarah in Fig. 3.13. Although these MCS cannot be resolved by current observing systems in most ocean basins, they can be recognised by their consistent convective signature over a period of up to several days. The MCS interaction with the cyclone can be analysed using the binary cyclone method in Fig. 3.9. In this case, the location of the MCS is taken as either the geometric centre of the cloud mass, or the centre of circulation, when one is evident.

Recent work by Carr (personal communication, 1993) indicates that sharp changes of tropical cyclone motion may arise from cyclones interacting with nearby synoptic scale systems. For example, in the western North Pacific it is common for cyclones to develop on the eastern flank of the major monsoon low in the Philippine Sea. These cyclones track around the low, but may also merge and coalesce with. After merger, the combined cyclone and monsoon low remnants will turn sharply poleward. One example of this process is the final sharp turn back on a poleward track by Typhoon Sarah (Fig. 3.13). In this case the movement seems to have been the result of a complex interaction and merger of the typhoon, a mesoscale complex, and the large monsoon low.

Figure 3.13: Meandering track of Typhoon Sarah (1989) during interaction with mesoscale convective system Alpha (Holland and Lander, 1993). Also shown is the nearly straight track of the geometric centre of the two systems and their cyclonic orbit about this centre (inset).

Recommendation: These interactions between tropical cyclones and surrounding weather systems is subject to intense current research. We expect that such research will lead to significant improvements in our knowledge with improved conceptual models for forecasting. In the meantime, we strongly recommend that care be taken to identify nearby systems that have the potential for introducing erratic changes in cyclone movement.

In particular, whenever, two tropical cyclones occur within 2000 km of each other, or there is evidence of a distinct mesoscale convective system within 1000 km of a tropical cyclone, their centroid relative movement should be plotted. This plot will provide valuable information on both the degree and type of interaction relative to the model in Fig. 3.12. Once binary orbit has commenced, the motion of the centroid should be analysed. The total cyclone motion is then forecast as a combination of binary orbit and centroid motion.

Hallin (1993) has developed a computer program for use at JTWC that tracks two adjacent tropical cyclones, indicates capture, rotation and escape, calculates the centroid of the rotation, builds a forecast of the centroid motion using CLIPER, then combines all parameters to produce track forecasts for each cyclone. The technique is being expanded to utilise numerical model and statistical forecast guidance.

3.3.6 Satellite Techniques

Satellite techniques involve interpreting cloud patterns that are associated with specific types of motion, or changes in motion. In contrast to the widespread adoption of the Dvorak method for estimating cyclone intensity, satellite based motion techniques have not gained wide acceptance. Those that have been developed usually have related movement to the orientation of outer convective bands, or to changes in shape of the inner cloud mass. They have not been objectively verified under operational conditions and should be used with some caution.

There is a tendency for tropical cyclones to move towards the downstream end of convective cloud bands in the outer circulation (Lajoie and Nicholls, 1974; Lajoie, 1976a). Changes in the orientation of such cloud bands indicate that a similar change in cyclone direction may occur in the next 12-24 hours. Further, tropical cyclones do not continue towards, nor curve towards, any cumulonimbus free sector in the outer circulation. Fett and Brand (1975) noted that rotation of gross cloud features (such as an elliptical cloud mass, or a major outer band) provide a very good indication of cyclone direction changes in the next 24 hours. The rule of thumb is to turn the cyclone to a constant heading relative to the rotated cloud mass. A mass of middle-level cloud streaming poleward from a tropical cyclone indicates potential recurvature (Ramage, 1974), and disappearance of this cloud indicates that a more westward track is likely.

Substantial qualitative information can be obtained form use of observations in the water-vapour emission window of 6.7 µm (Velden, 1987). By clearly indicating dry and moist regions, such imagery can provide excellent information on the location and movement of synoptic features likely to influence the storm motion. A set of pattern-recognition techniques for interpreting such imagery were proposed by Dvorak (1984), but these have not been operationally verified.

Animated satellite data have proven valuable in allowing forecasters to track the movement of peripheral synoptic features that influence tropical cyclone motion. It is recommended that only the last 24 hours be assessed for trends in movement of these peripheral features.

3.3.7 Expert Systems

The term "Expert System" covers a variety of methods. The check sheets used in many forecast offices to proceed logically through a specific forecast process are a rudimentary form of expert system. More complex decision trees also have been developed for interactive use on workstations or personal computers. For example, the synoptic classification of Hanstrum and Foley ( Fig. 3.8) can be applied by an interactive program on a personal computer, which takes the forecaster logically through a series of questions to determine the relevant forecast situation. Software now exists for automated development of decision trees, which present the answer in a probabilistic form, rather than the yes/no format of traditional check sheets.

TAPT (Typhoon Acceleration Prediction Technique; Wier, 1982; JTWC, 1992) is used at the JTWC. This technique uses surrounding wind fields to estimate the potential for rapid or delayed acceleration associated with poleward oriented or recurving tropical cyclones. Guidelines are provided for duration of acceleration, maximum acceleration and typhoon path.

Other forms of expert system have not yet been applied to tropical cyclone motion forecasting, so only a general description of potential approaches is provided here. One area of considerable potential for meteorological applications is the use of visualisation techniques. Work being undertaken in the biological sciences indicates that specific patterns, for example on satellite imagery, could be recognised objectively and entered into decision trees. Thus, a sequence of satellite imagery and cyclone tracks could be used to develop new motion forecast algorithms.

Greater complexity can be achieved by the use of high order expert systems, such as neural networks. These accept a set of input data and results from historical events, such as past tropical cyclone tracks, then iteratively calculate all possible paths via a series of intermediate points, known as nodes. Advantages include the potential to recognise extreme cases and branch points. They require massive computing facilities to develop, however, and have not been shown to be better than standard statistical techniques.

Some limited research is being undertaken to document the approach made by experienced forecasters with the aim of reproducing these in an expert system. However, thus far forecaster's thought processes have successfully defied logical description!

3.3.8 Dynamical Methods

The family of dynamical forecast methods extends from simple, quasi-analytical trajectory techniques to complex numerical models integrated over many days on a global domain. A summary of current techniques is presented in Table 3.3. Detailed descriptions of the major numerical models are provided in Chapter 8 and a comprehensive survey may be found in Elsberry (1993). Surveys of earlier models have been made by Elsberry (1979, 1987) and in Anthes (1982).

The BAM (Beta and Advection Model) and FBAM (FNOC BAM) are used at NHC and JTWC, respectively. These techniques were developed from the analytic motion equations in Holland (1984) by Marks (1989). The forecasts from a global spectral model are filtered to remove small scales and averaged through either the 850-300 hPa or 850-200 hPa layer to provide an advecting flow, to which is added a first-order approximation of the beta effect. Although very simple in its application and dynamics, the model is competitive with other dynamical techniques and consistently improves upon CLIPER (JTWC Annual Tropical Cyclone Report Series; DeMaria et al., 1990).

The National Hurricane Center replaced the SANBAR model (Sanders and Burpee, 1968) with VICBAR in 1990. Recent statistics provided by Lawrence (personal communication, 1993) indicate that VICBAR is competitive with the best forecast techniques to 24-36 hours.

The Dynamical Space Mean model is used at the Hong Kong Royal Observatory and assumes that the cyclone can be considered to be an air parcel moving under the influence of the large scale steering and pressure gradients. These are derived using four grid-points surrounding the storm. Persistence is incorporated through the past 12 hour movement. According to WMO (1987) this technique provides competitive 24 hour forecasts, but caution is recommended with its use until local statistics have been obtained.

The BARO model was developed in BMRC for interactive use on a personal computer in the TCM-90 experiment (Holland et al., 1991; Table 3.4), during which it provided valuable forecast guidance. It has subsequently been adapted both for operational use in the Australian region and for generic use on workstations in any tropical cyclone region. It utilises the non-divergent barotropic vorticity equation, a Barnes analysis and a relocatable grid to facilitate interactive operational use on any atmospheric layer mean. A modified form of the analytic wind profile of Holland (1980) is used to provide a bogus cyclone at the analysis stage.

OTCM (One-way influence Tropical Cyclone Model; Hodur and Burke, 1978) is a coarse resolution (205 km) grid-point model with three vertical levels used by the JTWC. It is integrated on a 6400x4700 km domain centred on the cyclone, is initialised by 6 or 12 hour prognostic fields from NOGAPS, and uses NOGAPS forecasts to update the boundaries. The initial fields are smoothed in the vicinity of the storm, a persistence-of-track steering component and a symmetric bogus are then added. This vortex is maintained against dissipation by an analytic heating function.

Statistics provided by the JTWC in their Annual Tropical Cyclone Reports, indicate that the OTCM is amongst the best techniques. Both NOGAPS and OTCM perform substantially better than CLIPER at periods greater than 24 hours.

3.3.9 Statistical and Statistical/Dynamical Techniques

Statistical and statistical/dynamical techniques are developed from some form of statistical screening using past storms and a variety of input information. Statistical/Dynamical methods utilise output from a numerical model. Full details of the procedures used may be found in Elsberry (1987). WMO (1979), and Bureau of Meteorology (1978) contain surveys of past techniques, whilst Neumann and Pelissier (1981a,b) provide an operational evaluation. A summary of current techniques is provided in Table 3.5.

The first three statistical techniques use a grid located on the tropical cyclone to develop predictors for use on current analyses, together with past cyclone motion. Climatology is included in NHC72 by CLIPER and in TOPEND by CYCLOGUE. The Vietnam technique uses an empirical orthogonal expansion of the mid-level geopotential and surface pressure fields, combined with empirical parameters related to surface pressure gradient and past cyclone motion.

The statistical/dynamical techniques use forecasts from a numerical model as predictors. CSUM is used by JTWC and uses cyclone positions and 500 hPa height fields from the NOGAPS analyses over the past 24 hours, together with NOGAPS prognoses at 24 and 48 hours. For intense cyclones following recurvature, 200 hPa height data are used instead of those at 500 hPa. NHC90 is an upgrade of NHC83 and is used by NHC in the North Atlantic; JTWC92 is NHC90 applied to western North Pacific cyclones. SD75 was developed in China and is similar to the dynamical space mean technique (Table 3.3).

CSUM was developed using actual analyses instead of prognoses. Often referred to as a "perfect prognosis" approach, this means that CSUM is not optimised for a particular version of a numerical model. But it also does not degrade as the model changes, and, in theory, CSUM should improve as the numerical forecasts become more accurate. CSUM also has the unique feature of using three different sets of regression equations, depending on whether the cyclone is equatorwards of, on, or poleward of the subtropical ridge. This technique has consistently been better than CLIPER in the western North Pacific, where it is the best statistical technique (JTWC Annual Tropical Cyclone Reports).

The major advance made with NHC83 was to rotate the analysis and forecast grids to a Lagrangian coordinate system centred on the storm and oriented along its track. This decouples the along- and cross-track errors and produces an optimal statistical result (Shapiro and Neumann, 1984). NHC83 has consistently been the best forecast aid past 24 hours at NHC and has shown significant potential in trials at JTWC.

3.3.10 Strike Probability Forecasts

One method of objectively defining the uncertainty inherent in cyclone forecasting is to utilise strike probabilities (Jarrell, 1978). Probabilistic forecasts provide valuable early guidance in estimating the risk of tropical cyclones effecting important, or vulnerable facilities. Operational methods include those of Neumann (1987) and Templeton and Keenan (1982).

Strike probabilities can be presented as either a set of probability contours (Fig. 3.14) or a time series of probability of a particular region (such as a major industrial complex) being affected. The probabilities are derived from a knowledge of past cyclones and forecast errors in the region of interest (see Neumann, 1987). New programs aimed at objectively estimating the uncertainty in dynamical forecast models (Leslie and Holland, 1991) should help to identify critical situations when major forecast errors are possible.

The use of probability forecasts can be highly misleading to the public. In particular, the size of the strike probability is a function of time along the predicted track and the probability of across-track error. Thus, for example, a 36 h motion along a confident track will result in lower probabilities of landfall than will a highly improbable track from a cyclone 12 h from landfall. Care and good public educations programs must accompany their introduction.

Figure 3.14: 24 hour Strike probability contours (at 10% spacing) for Tropical Cyclone Joy off the Australian east coast (Woodcock, personal communication, 1993).

An advantage of strike probabilities is that users can make objective decisions based on potential costs and losses. This method, often referred to as "cost-loss" decision making is based on the simple formula:

      (3.4)

where C is the cost of taking action to avoid cyclone effects and L is the loss arising from taking no action. From an entirely economic viewpoint, action does not commence until CL is less than unity. More complex and well founded methods can be derived for specific circumstances.

1. Known as a Marsden square

2. These annual frequencies may differ from those in Table 1.3 due to the different periods of record.

Contents Chapter 3.4